This document compares computational fluid dynamics (CFD) modeling and physical flow modeling. Both methods can accurately simulate fluid flow behavior but have advantages and disadvantages. CFD models are usually faster and less expensive than physical models. However, physical models are better for simulating some phenomena like particulate behavior. For complex projects, using both CFD and physical modeling may provide the most comprehensive results. The appropriate modeling method depends on the specific application and objectives.

1. Pros and Cons
of CFD and Physical Flow Modeling
A White Paper by:
Kevin W. Linfield, Ph.D., P.E.
Robert G. Mudry, P.E.
August, 2008
© 2008 Airflow Sciences Corporation
All Rights Reserved

2. In a CFD model, the three-dimensional domain is
built in the computer via a CAD model. Acomputational
mesh is then inserted into the domain – this mesh
divides the region where flow travels into many control
volumes, or cells. It is not uncommon for a CFD model
to contain millions of these cells. The software then
solves the equations of fluid motion (Conservation of
Mass, Momentum, and Energy) in every one of these
cells. The results are plotted as color contours to depict
the flow parameters at any location within the domain.
Thus, it is possible to analyze millions of velocities,
pressures, temperatures, species concentrations,
and other values. Computer-generated animations
can also be created that provide flow visualization to
observe the “real-time” motion of the flows.
Figure 2 - CFD mesh for an electro-
static precipitator
Computational Fluid Dynamics (CFD) is a method of simulating fluid flow
behavior using high speed computers. There are well-known mathematical equations
that define how air and gases behave (Conservation of Mass, Momentum, and Energy).
These equations are extremely
complex (differential equations),
and thus can not be solved by
hand calculations except for very
simple geometries such as flow
around a cylinder. As computer
power increased in the 1970s,
the aerospace industry led the
way in developing software to
approximate solutions to these
equations for complicated
flows around air and space craft. Over the past few decades, these software tools
have advanced to a point where accurate solutions can be obtained for complex flows,
including heat transfer, particle tracking, and chemical reactions.
When it comes to flow modeling to optimize
performance or to develop solutions for flow-related
problems, a frequent question that industry engineers
ask is “Which is better – a CFD or physical (scale)
flow model?”. The short answer is “It Depends”.
Figure 1 – CFD was first developed for the aerospace industry
NASA
Background

3. Once the physical model is constructed, large
fans are used to draw air through the model at a flow
rate that provides similar fluid dynamic behavior to the
full scale system. Flow characteristics are measured
over a grid of traverse points with an inserted probe.
Values for velocity and pressure at select locations
are thus obtained. Dust can be injected into a model
to simulate the behavior of particulate in a system (to
assess ash deposition, for example). Of course, the
model is constructed with clear walls or windows so that
flow patterns can be observed via smoke flow, strings, or bubbles. Model results can be
presented as color contours, histograms, or other plotting methods similar to field testing.
It is difficult to determine how long physical
flow modeling has been used in engineering
applications. Obviously, full-scale versions of land
and sea vessels were tested via trial-and-error for
centuries to optimize designs. In the early 1900s,
the Wright Brothers tested a scaled version of an
airfoil in a small wind tunnel that led to the age of
flight. Since the 1960s, scale models have been
used to assess flow patterns in power plant duct
systems, pollution control equipment, and boilers.
Today, many of these models are built to a scale of
1:8 to 1:16, with 1:12 being a common scale factor.
With either type of model, the flow patterns through the system are quantified and
the model geometry is iteratively altered in order to optimize the flow. The location and
shape of control devices such as turning vanes, mixers, baffles, and dampers are thus
determined such that the design objectives are attained.
Figure 5. Smoke flow through an
SCR physical model
Figure 3. Wright Brother’s wind tunnel
Figure 4. Physical flow model
of a power plant dry scrubber
and baghouse system
Accuracy
With the proliferation of high speed computers, the resolution and cell size of
CFD models has improved dramatically over the past few decades. Airflow Sciences
Corporation, which has used both modeling methods since 1975, has made numerous
comparisons between CFD modeling, physical modeling, and field testing. Results
indicate that both types of models share the same accuracy when it comes to velocities
and pressures. For more on this please visit ASC’s web site (www.airflowsciences.
com) for Conference Proceedings which make this comparison with respect to ESP and
scrubber modeling.
www.wright-brothers.org

4. There are certain areas where CFD and physical model results differ and it is
not clear which provides the best real-world results. For instance, in SCR modeling,
CFD models tend to predict slightly worse ammonia uniformity at the catalyst compared
to physical models. Industry comfort is with the physical model in this case, and it is
possible that the underlying mesh is not fine enough to resolve all the details of the
injection and mixing. That said, there is not a lot of specific data published that shows
how well either model matches real-world test data.
Similarly, for wet FGD absorbers and stacks, physical models are often used
with liquid water injected into the models. Though the droplet size is not scaled properly,
and evaporation is not represented accurately, some industry designers find value in the
results and utilize their experience to interpret the results of the wet modeling. This is
a very complex flow phenomenon, where two-phase flow momentum effects the droplet
agglomeration exist. This is equally difficult to simulate with a CFD model, even with
evaporation and thermal effects simulated. So both model types have drawbacks and
industry experience in applying the results to the real world become important.
CFD model studies are generally 20-40% less than a comparable physical
model effort. This is tied quite strongly to the labor difference in model construction that
influences the schedule. Also, many CFD tasks can be automated with the computer,
including the design optimization process, whereas these tasks are primarily manual
with the physical model.
CFD modeling is almost always faster than physical modeling. In many cases,
design results from a CFD model are available several weeks before similar results from
a scale model. And the more complicated or repetitive the model geometry is, the more
advantage the CFD model has. This has to do with three factors: 1) the CFD mesh can
usually be built faster than a scale model can be fabricated, 2) for repetitive or symmetric
duct systems, portions of a CFD model can be copied and pasted while all pieces of the
physical model need to be built separately, and 3) once a CFD model is built, it can be
run simultaneously on separate computers. Thus, several designs can be evaluated at
the same time, while only one physical model exists to evaluate designs.
Figure 6. Comparison of CFD and physical model results for an FGD duct system
where flow from 3 units (1,750 MW total) combine to feed 3 new booster fans (CFD
pressure drop 1.19 IWC; Physical pressure drop 1.27 IWC)
Schedule
Modeling Cost

5. Most physical models are built to scale, typically 1:12 or 1:16 for power plant
models. CFD models are almost always built full size (1:1 scale). Care must be taken
in computer models to ensure that the correct number, size, and shape of computational
cells are used, and the level of detail to include must be considered in a scaled model to
ensure geometric and dynamic similarity is maintained. In a CFD model, the Reynolds
Number is often matched exactly, while in a physical model industry generally tries to
match the Reynolds Number regime (i.e., laminar or turbulent). Both are fine as long
as the boundary layer is negligible. This is generally the case for large power plant duct
systems. Note, however, that one must closely match the exact value of the Reynolds
Number if the objective is to determine lift or drag characteristics, or any system where
the boundary layer along a surface is important.
Figure 7. Comparison of CFD and physical model of a windbox
In general, solid particle drop-out or re-entrainment is more accurate in a physical
model. These tests help assess whether particulate (such as flyash) will fall out of the gas
stream at lower unit flow rates. It is important to run the
physical model at comparable velocities to the actual
system, taking into account particulate aerodynamic
characteristics which can be determined via wind tunnel
tests. CFD results can be used to assess potential
areas for particulate drop-out by examining low velocity
regions near duct floors and other surfaces, but CFD
cannot yet predict re-entrainment of particles as the
system flow rate ramps up. This is because particulate
build-up and re-entrainment are time-dependent
phenomena. A physical model can be used to observe
the particle behavior over time, but a CFD model is
generally run as a steady-state simulation.
Figure 8. Physical model dust
testing (dust accumulation simulated
with fine white powder)
Scale
Particulate

6. Simulation of a chemical reaction (such as combustion or change-of-state) can
realistically only be done with a computational model or a laboratory test that includes the
reactions. The latter would not really be referred to by industry as a “physical flow model” as
much as a lab test (such as a combustion test chamber). Short of such a lab test, computer
flow modeling can be used to simulate complex processes, incorporating individual species
and compounds via reaction equations. Furnace combustion models are done via CFD
to assess items such as burner/OFA systems, NOx creation, gas temperature uniformity,
SNCR performance, slagging, and corrosion. Also, evaporative processes can only be fully
simulated in a CFD model due to the changes in temperature and the moisture transfer from
one state to another.
For complex temperature problems (especially those involving conduction, convection,
or radiation), CFD is really the only option. Physical models are often called “cold-flow models”
since room-temperature air is drawn through the domain. Methods have been devised to
simulate thermal mixing in a physical model (such as the merging of gas streams of differing
temperature) via an injected tracer gas. Unless they are run at temperature, however, physical
models cannot simulate heat transfer, addition of heat, or similar phenomena. CFD models
are run at the correct temperature, and take into account changes in density, viscosity, thermal
conductivity, and the heat transfer coefficient. CFD models of boiler combustion processes,
heat exchangers, and evaporative processes are thus possible.
Figure 10. CFD modeling of thermal mixing (SCR
inlet duct with economizer bypass flow)
Particulate tracking is often desired to assess items
such as Large Particle Ash pluggage, activated carbon/sorbent
injection, or flyash erosion issues. Particles “in flight” are better
simulated in a CFD model. This is because the CFD model is run
full scale and can thus match all the important factors for particle
behavior simultaneously (gravity, particle drag, gas velocity, gas
viscosity, particle Reynolds number, particle mass and size).
Some qualitative assessments of particle behavior “in flight” can
be performed with physical models, but because all the scale
factors and fluid dynamic properties are challenging to match
simultaneously, quantifiable results are more difficult to obtain.
Figure 9. CFD tracking of ash particles
in flight to assess LPA screen capture
Figure 11. Physical model testing of ammonia injec-
tion in an SCR via tracer gas simulation
586.0 602.8 619.6 636.4 653.2 670.0
Temperature (f)
Chemical
Reaction
Heat Transfer

7. Both types of models rely on color contour plots
and flow statistics (uniformity, min/max values, etc.) to
quantify results. Smoke injections and string tufts are
also used to visualize the flow field inside a scale model.
These are videotaped and photographed to document the
flow patterns. Dust testing results are also videotaped so
observations of particulate drop-out and re-entrainment
can be documented. Flow animations from CFD results
can provide similar views on the motion of the flow as a
physical model smoke test. CFD animations can also
present characteristics that are difficult to quantify in a
physical model (i.e., a visual tracking of injected gas molecules, such as SO3 or NH3, through
a duct).
Figure 12. Smoke flow details in a
physical model
Figure 13. CFD injection of activated
carbon upstream of an electrostatic
precipitator
(Left – full ESP; Right – close up at
lance location)
As noted above, there are certain flow characteristics that are best simulated with
a particular type of model. Since there are advantages and disadvantages of both models,
a number of new systems, particularly the more expensive pollution control devices such
as SCR and FGD, utilize both modeling methods to get the optimal design. For ductwork
systems, ESPs, or fabric filters, both methods have shown they offer similar results and ac-
ceptable designs; in these cases, the selection of the method often comes down to personal
preference of the OEM or the end user.
CFD models are usually stored on tape, CD-ROMS or DVDs which typically have a
much longer storage life and negligible space requirements. Physical models can take up
considerable space in a warehouse. A benefit of the physical model after the design effort is
that it can be used for other purposes, including as a training tool for plant staff or as a display
item for a plant lobby.
Seeing and touching a laboratory model can be more satisfying than looking at color
contour plots and animations of a virtual model. Many clients appreciate walking around a
3-D scale model and examining flow details around the vanes, through perforated plates, and
near internal structure. What’s best depends on personal preference.
Visualization
Conclusion
Storage
Touch & Feel

8. Corporate Office
12190 Hubbard Street
Livonia, MI 48150-1737 USA
Tel. (734) 525-0300
Fax (734) 525-0303
www.airflowsciences.com
Western Region Office
PO Box 22637
Carmel, CA 93922-0637 USA
Tel. (831) 624-8700
Southern Region Office
3709 Foster Hill Drive North
St. Petersburg, FL 33704-1140 USA
Tel. (727) 526-9805
Southeast Asia Agent
HANA Evertech Co. Ltd.
Jeff Jang
jjang@airflowsciences.com
+82-31-777-3780
www.airflowsciences.com
Copyright © 2008 Airflow Sciences Corporation
All rights reserved. No part of this publication may be
reproduced, stored in a retrieval system, or transmitted,
in any form of by any means, electronic, mechanical,
photocopying, recording, or otherwise, without the prior
permission of the copyright owners.
Any comments relating to the material contained in this
document may be submitted to asc@airflowsciences.com.